Does Speech Comprehension Require Phonemes?

This chapter reviews the arguments that have been raised against the unit-based account and presents some of the wide range of approaches that have been taken to modelling speech which do not involve speech units. Introductions to phonetics demonstrate that the speech signal consists of fluctuations in energy that occur over time in a wide range of frequencies. The chapter presents the evidence and arguments against the combinatorial approach. A small set of phonemes or distinctive features seemingly provides a way to tame the wild, complex, and unpredictable nature of the continuous speech signal. Proponents of distributional models argue that what is learned is the statistical distribution of phonetic cues. Linguistic representations and processes emerge through a predictive, error-driven discrimination process and therefore adapt to surprising events or information. Most studies on speech perception and word recognition have focused on the phonetic domain.